huggingface_downloader

huggingface_downloader is a native Dart utility for downloading complete model snapshots directly from Hugging Face Hub.

It provides a lightweight implementation similar in spirit to Python's huggingface_hub.snapshot_download(), allowing Dart applications, AI runtimes, and CLI tools to fetch Hugging Face repositories without external Python dependencies.

Ideal for:

• 🤖 downloading LLM repositories
• 🧠 fetching tokenizer/config/model artifacts
• ⚙️ automation pipelines
• 🧪 CI model fixture downloads
• 📦 building local inference environments
• 🔄 resumable large file downloads

---

Features

• 📥 Download complete Hugging Face repository snapshots
• 📂 Preserve nested directory structure
• 🔁 Resume interrupted downloads automatically
• ♻️ Optional full overwrite/redownload mode
• 🔐 Support gated/private repositories via HF token
• 🎯 Inclusion filters (--ext)
• 🚫 Exclusion filters (--exclude)
• 🤖 Built-in --llm-only mode for common LLM artifacts
• 📄 Returns downloaded file list programmatically
• 💻 CLI utility included
• 🧩 Native dependency-free Dart implementation using HttpClient

---

CLI Usage

Activate globally:

dart pub global activate huggingface_downloader

Run:

huggingface_downloader <repoId> <outputDir> [options]

---

Arguments

• <repoId> → Hugging Face repository id
• <outputDir> → local directory where files will be downloaded

---

Options

• --token=hf_xxx → Hugging Face access token for private/gated models
• --ext=.json,.safetensors → download only selected extensions
• --exclude=.onnx,.gguf,.pt → exclude unwanted artifact types
• --llm-only → keep only common LLM files
• --overwrite → force full redownload even if files already exist
• -h, --help → show help message

---

CLI Examples

Download SmolLM2:

huggingface_downloader HuggingFaceTB/SmolLM2-135M-Instruct ./models/smollm2 --llm-only

Force overwrite existing local files:

huggingface_downloader HuggingFaceTB/SmolLM2-135M-Instruct ./models/smollm2 --llm-only --overwrite

Download Qwen excluding ONNX/GGUF:

huggingface_downloader Qwen/Qwen2-0.5B ./models/qwen2 --exclude=.onnx,.gguf

Download private/gated repository:

huggingface_downloader meta-llama/Llama-3.2-1B-Instruct ./models/llama --token=hf_xxxxxxxxx --llm-only

---

Example CLI Output

HuggingFace Downloader
----------------------
Repository : HuggingFaceTB/SmolLM2-135M-Instruct
Output Dir : ./models/smollm2
Mode       : LLM ONLY

Files selected for download: 6
  config.json
  tokenizer.json
  tokenizer_config.json
  generation_config.json
  special_tokens_map.json
  model.safetensors

model.safetensors -> 100.0% (542.13 MB / 542.13 MB)

Download completed successfully.

Downloaded files (6):
----------------------
models/smollm2/config.json
models/smollm2/tokenizer.json
models/smollm2/tokenizer_config.json
models/smollm2/generation_config.json
models/smollm2/special_tokens_map.json
models/smollm2/model.safetensors

---

Programmatic Usage

huggingface_downloader can be embedded directly into Dart scripts, model preparation tools, or local inference pipelines.

import 'dart:io';
import 'package:huggingface_downloader/huggingface_downloader.dart';

Future<void> main() async {
  final downloader = HuggingFaceDownloader();

  final files = await downloader.downloadSnapshot(
    repoId: 'HuggingFaceTB/SmolLM2-135M-Instruct',
    localDir: Directory('./models/smollm2'),
    excludeExtensions: ['.onnx', '.gguf'],
    overwriteExisting: false,
    progress: (file, received, total) {
      print('$file -> $received / $total');
    },
  );

  for (final file in files) {
    print('Downloaded: ${file.path}');
  }

  downloader.close();
}

---

LLM Only Mode

--llm-only is a convenience mode designed for the most common local inference workflow.

Automatically includes:

• .json
• .txt
• .model
• .safetensors
• .bin

Automatically excludes:

• .onnx
• .gguf
• .h5
• .msgpack
• .tflite
• .pt
• .pth
• .ot
• .ckpt

This avoids downloading unrelated framework artifacts commonly present in Hugging Face repositories.

---

How It Works

1. Fetch repository manifest from Hugging Face Hub API
2. Read repository file list (siblings)
3. Apply include/exclude filters
4. Resolve each file through Hugging Face resolve endpoints
5. Resume or overwrite existing files as requested
6. Stream files directly to disk
7. Return the final downloaded file list

---

Why This Package

Python has huggingface_hub.snapshot_download().

Dart had no native equivalent.

huggingface_downloader fills that gap with a simple automation-friendly downloader suitable for:

• Dart AI runtimes
• local LLM preparation
• CI artifact fetching
• reproducible model bootstrapping
• CLI tooling

---

Test Repository

For automated tests and CI validation, a tiny public repository works very well:

fxmarty/really-tiny-falcon-testing

---

Issues & Feature Requests

Please report issues or request features via the issue tracker.

---

Author

Graciliano M. Passos: gmpassos@GitHub

---

License

Dart free & open-source - BSD 3-Clause License.